A recruitment agency that has been trading for ten years is sitting on something strange: tens of thousands of people it knows, and a working memory of a few hundred.
Every one of those records was expensive. Someone sourced them, spoke to them, took them through a process. Many were placed. Many more got to final stage and lost to another candidate, which means they were good enough to be shortlisted by a client who was paying attention.
Then a role closes, and they go quiet. Not deleted, not lost. Just never surfaced again, because nothing in the system will raise them at the moment they are relevant.
Why the database goes still
The honest reason is not laziness and it is not bad software exactly. It is that recall in most systems is a query, and a query requires you to already suspect the answer.
To find the finance candidate you interviewed two years ago, you have to remember that they exist, guess how their job title was recorded, and know which of your four spellings of their skill set was used that week. Keyword search rewards people who remember. That is precisely the thing a busy desk cannot do at volume.
So the search happens at the surface. You look at the last two months of applicants, because those you remember. The other twenty-nine thousand records are functionally invisible, and the agency goes back to the job boards to buy attention it already owns.
The economics are worse than they look
Consider what happens when a role lands and the desk goes external immediately.
- You pay again for reach you have already paid for, in advertising or in a sourcing subscription.
- You approach people with no relationship, so your response rate is the market's, not yours.
- You lose the speed advantage entirely. A candidate you placed three years ago will take your call today. A cold candidate will take three weeks to decide whether to reply.
- And someone in your own database, who already knows you and is currently open to a move, never hears about the role at all.
The last point is the one that should sting. It is not a missed efficiency. It is a placement that was available and did not happen.
Keyword search rewards the recruiter who remembers. That is exactly the thing a busy desk cannot do at volume.
Retrieval is the actual problem
The fix is not more data. It is making the data you own respond to a role rather than to a query.
That means the system reads a job description the way a recruiter does, as a shape rather than as a bag of keywords, and matches against what it understands about each person: what they have actually done, how they have moved, where they sit relative to the brief. It means the ranking runs when the role is created, not when someone thinks to search.
And it means being explicit about fit rather than pretending to certainty. A shortlist that says strong, moderate and stretch, relative to the best available candidate for that specific role, is more useful than a percentage that implies a precision nobody has. A recruiter can act on a stretch candidate. They cannot act on 71%.
What changes on the desk
When retrieval works, the first move on a new role stops being advertising and starts being a shortlist from your own book, ready before you have finished reading the brief.
Some of those people will be wrong, and cutting them is not wasted effort: each rejection is a statement about how this desk judges, and a system that keeps them gets closer to your judgement over time. What does not happen is the twenty-nine thousand records staying invisible for another year.
Most agencies do not need a bigger database. They need the one they have to answer when a role lands.